Diagnosis and procedure data analysis - emr package in R

• Provide mechanisms to integrate, analyze, and visualize diagnosis and procedure records
• Simplifie and accelerate the process of extracting ready-for-analysis datasets from EHRs

• An R package
• Open-source software on GitHub

R : Core (engine) Source

RStudio : IDE (dashboard) Source

R : Core (iPhone)

• How to install packages?

install.packages("Package name")

• How to load the installed packages?

library(Package name)

• Basic functions
• Functions provided by packages (such as emr)
• function_name(argument1, argument2,....)

For example, we can get the average of numbers by mean() function:

mean(c(1,2,3,4,5,6)) ##average of 1~6

[1] 3.5

If you don't know the defined order, you can assign the arguments by name

seq(from=1,to=9,by=2)

[1] 1 3 5 7 9

• Not on CRAN right now
• Install from GitHub (by devtools package)

install.packages("devtools")
devtools::install_github("DHLab-CGU/emr")
library(emr)

install.packages("tidyverse")
install.packages("tableone")
library(tidyverse)
library(tableone)

Explore the comorbidity or underlying diseases difference between patent ductus arteriosus (PDA) and the non-PDA patients

• Sample file sampleFile_MIMIC
• MIMIC-III, a dataset of 7,833 newborn patients with 45,674 admissions

head(sample_MIMICIII)

1. Select PDA case
2. Group diagnosis records
3. Visualize the diagnosis records
4. Generate “Table 1” for the analysis

• Split the newborns (n = 7,833) into case (PDA) and control (non-PDA) groups:
  • PDA: 381
  • non-PDA: 7452

selectedPDA <- 
  selectCases(sample_MIMICIII, 
              SUBJECT_ID, ICD9_CODE, 
              ADMITTIME, "2015/10/01", 
              ICD, caseCondition = "^7470", 
              caseCount = 1, 
              CaseName = "PDA")

selectedPDA

• DxDataFile: sample_MIMICIII
• idColName: SUBJECT_ID
• icdColName: ICD9_CODE
• dateColName: ADMITTIME

selectedPDA <- 
  selectCases(sample_MIMICIII, 
              SUBJECT_ID, ICD9_CODE, 
              ADMITTIME, "2015/10/01", 
              ICD, caseCondition = "^7470", 
              caseCount = 1, 
              CaseName = "PDA")

• icd10usingDate: 2015/10/01
• groupDataType: ICD (CCS, AHRQ, PheWAS…)
• caseCount: 1
• CaseName = “PDA”

selectedPDA <- 
  selectCases(sample_MIMICIII, 
              SUBJECT_ID, ICD9_CODE, 
              ADMITTIME, "2015/10/01", 
              ICD, caseCondition = "^7470", 
              caseCount = 1, 
              CaseName = "PDA")

selectedPDA

sample_MIMICIII[SUBJECT_ID=="32433"&
                        grepl("^747",ICD9_CODE)]

sample_MIMICIII[1:6,
                c("SUBJECT_ID", 
                  "ICD9_CODE", "ADMITTIME")]

• Group diagnostic codes with Elixhauser comorbidity
• Convert the long format of data into a wide format
• Arguments:
  • groupDataType: Elix (AHRQ, CCS)
  • isDescription: F (T)
  • selectedCaseFile: selectedPDA (From case select step)

elix <- 
  groupedDataLongToWide(
    sample_MIMICIII,SUBJECT_ID,
    ICD9_CODE,ADMITTIME, 
    "9999-10-01", groupDataType = Elix, 
    isDescription = F, 
    selectedCaseFile = selectedPDA)

elix

plot_Elixhauser <- plot_groupedData(elix)
plot_Elixhauser$graph

plot_Elixhauser$sigCate

var_col <- 
  names(elix)[2:(length(elix)-1)]
table_Elix <- 
  CreateTableOne(vars = var_col,
                 strata = "selectedCase", 
                 data = as.data.frame(elix), 
                 factorVars = var_col)
table_Elix

1. Code standardization
2. Data integration
3. Exploratory data analysis (EDA) preparation
4. Visualization

• For illustrating the main functions of emr package
• Simulated medical data set of 38 patients with 300 records

sampleDxFile[6:10]

• Standardize the ICD codes into a uniform format before analysis
  • short format: xxxyy
  • decimal format: xxx.yy

• The formats of ICD used in the grouping tables could be different
• Already integrated into grouping method
  

• Transform the diagnosis codes into uniform format before analysis
• Check the correctness of ICD codes: with clear warning messages
  • Help researchers identify the potential coding error that may affect analysis

• Uniform code format functions:
  • Decimal format: IcdDxShortToDecimal
  • Short format: IcdDxDecimalToShort

decimal <- 
  IcdDxShortToDecimal(sampleDxFile, ICD, 
                      Date, "2015/10/01")
short <- 
  IcdDxDecimalToShort(sampleDxFile, ICD, 
                      Date, "2015/10/01")

cbind(decimal = decimal$ICD[6:10], 
      short = short$ICD[6:10])

• Warning message: identify the potential coding error
• There are two types of warning message:
  • wrong ICD code format
  • wrong ICD code version
• The suggestoin helps user adjust the incorrect ICD-9 code with unspecified classification (now only ICD-9 codes has suggestion).
  

decimal$Error

• We provide four strategies to group EHR diagnostic code.

• After grouping, this package provides functions for querying matching cases, splitting data and generating condition era

Integrated EHR data:

1. Code classification
2. Case selection
3. Patient record period
4. Data split
5. Condition era calculation

• Four strategies to group EHR diagnosis codes:
  • CCS
    • IcdDxToCCS()
    • IcdDxToCCSLvl()
  • PheWAS: IcdDxToPheWAS()
  • Comorbidity (Elixhauser, AHRQ, and Charlson): IcdDxToComorbid()
  • Customized grouping methods:
    • IcdDxToCustom()
    • IcdDxToCustomGrep()

• Grouped data by CCS multiple level 2

CCSlvl <- 
  IcdDxToCCSLvl(sampleDxFile, 
                ID, ICD, Date, 
                "2015-10-01", 2)

CCSlvl$groupedDT

• Summarised grouped data by CCS multiple level 2

CCSlvl$summarised_groupedDT

• Grouped data by CCS single level

CCS <- 
  IcdDxToCCS(sampleDxFile, 
                ID, ICD, Date, 
                "2015-10-01")

CCS$groupedDT

• Summarised grouped data by CCS single level

CCS$summarised_groupedDT

• Query the matched cases
• groupDataType: ccslvl2

case <- 
  selectCases(sampleDxFile, ID, ICD, Date, 
              "2015/10/01", ccslvl2, 
              caseCondition = "Diseases of the urinary system", 
              caseCount = 1, 
              CaseName = "Selected")

case[20:25]

• Get the first/last record date for each patient

recordDate <- 
  patientRecordDate(sampleDxFile, ID, ICD, Date)

recordDate

1. Select PDA case

1-1. Wash-out and follow-up period <—–

1. Group diagnosis records
2. Visualize the diagnosis records
3. Generate “Table 1” for the analysis

• Wash-out: period of records before the index date (for example, the first Diseases of the urinary system diagnosis)
• Follow-up: period of records after the index date

WFPeriod<-merge(case,recordDate,by="ID")

WFPeriod[,list(ID,firstCaseDate,endCaseDate,
                     firstRecordDate,endRecordDate)]

WFPeriod$Washout<-
  WFPeriod$firstCaseDate-WFPeriod$firstRecordDate
WFPeriod$Followup<-
  WFPeriod$endRecordDate-WFPeriod$firstCaseDate

WFPeriod[,list(ID,firstCaseDate,endCaseDate,
               firstRecordDate,endRecordDate,
               Washout,Followup)]

• Split data by the date of clinical event.
  • Identify the data is recorded before or after the clinical event.
  • Count the number of windows between the record date and index date.

1. Select PDA case

1-1. Wash-out and follow-up period

1. Group diagnosis records

2-1. Diagnosis before and after an important event <—–

1. Visualize the diagnosis records
2. Generate “Table 1” for the analysis

These records can be used for a given patient as an index date for data split function.

indexDateTable <- 
  data.frame(ID = c("A0","B0","C0","D0"), 
             indexDate = c("2023-08-12", 
                           "2024-02-12",
                           "2015-12-05", 
                           "2017-01-29"),
             stringsAsFactors = FALSE)

The important event, could be the first diagnosis date or treatment date

indexDateTable

• Gap: length of the window. Default set to 30.
• indexDateFile: indexDateTable

certainPatient <- 
  sampleDxFile[grepl("A0|B0|C0|D0",ID),]
splitedData <- 
  splitDataByDate(certainPatient, ID, ICD, Date, 
                  indexDateTable, Gap = 30) 

splitedData[15:19,]

Diagnosis 180 days before the index date

splitedData[timeTag=="B"&window<=6,]

• Convert the grouped data in long format into a wide format which is fit to others analytical and plotting function
• In the wide format, the variable can be in numeric and binary form
• One step method!

• DxDataFile: can be original data or splited and filtered data

wideData <- 
  groupedDataLongToWide(sampleDxFile, ID, 
                        ICD, Date, 
                        "2015-10-01", elix, 
                        isDescription = F)

wideData[,-(2:14))

• Visualized the overview of grouped clinical data
  • Pareto chart for error ICD list
  • Histogram plot for grouped data

• Plot for error ICD list
• In the first step, code standardization,the output include diagnosis codes with potential error

plotErrorICD <- 
  plot_errorICD(decimal$Error)
plotErrorICD$graph

• Top 10 ICD codes with errord

plotErrorICD$ICD

• Plot of diagnostic categories
• Summarizes the percentage of diagnostic categories in a histogram plot

plotGroupedData <- 
  plot_groupedData(wideData)
plotGroupedData$graph

Top 10 diagnostic categories grouped by Elixhauser comorbidity

plotGroupedData$sigCate

• Dataset: 953,294 unique patients with 7,948,418 diagnosis records.
  • Code standardization took 6.2 seconds
  • Data integration (including code standardization):
    • code classification with Charlson comorbidity took 8.5 seconds.
    • case selection took 14.3 seconds
  • EDA preparation took 16.5 seconds.
  • Visualizatoin: histogram plot took 1.0 seconds.

• Geting start
  • Diagnosis data
    • English: https://dhlab-cgu.github.io/emr/articles/Eng_Diagnosis.html
    • 中文: https://dhlab-cgu.github.io/emr/articles/Chi_Diagnosis.html

• ICD procedure code format

• ICD format in grouping methods

• Sample file for procedure record, which has 3 patients and 170 records.

samplePrFile[84:88]

• Uniform code format functions:
  • Decimal format: IcdPrShortToDecimal
  • Short format: IcdPrDecimalToShort
• Procedure code standardization also has warning messages helping users identify the potential coding error.

• Collapse ICD procedure codes into clinically meaningful categories for further analysis
• The functions of grouping procedure record:
  
  • CCS
    • The CCS classification with single and multiple levels
    • Similar to the diagnostic CCS classification.
  • Procedure class
    

ProcedureClass <- 
  IcdPrToProcedureClass(samplePrFile, ID, ICD, 
                        Date, "2015-10-01")

ProcedureClass$groupedDT

• Geting start
  • Diagnosis data
    • English: https://dhlab-cgu.github.io/emr/articles/Eng_Diagnosis.html
    • 中文: https://dhlab-cgu.github.io/emr/articles/Chi_Diagnosis.html
  • Procedure data
    • English: https://dhlab-cgu.github.io/emr/articles/Eng_Procedure.html
    • 中文: https://dhlab-cgu.github.io/emr/articles/Chi_Procedure.html